The complex environments of Measurement-While-Drilling (MWD), characterized by intense vibrations and high temperatures, can lead to random attenuation of optical intensity, phase mismatch, and signal distortion, thereby reducing the signal-to-noise ratio (SNR) of optical signals. Under severe channel fluctuations, forward stack decoding (FSD) tends to search more invalid nodes, increasing decoding complexity. This paper proposes an unquantized forward stack decoding algorithm that dynamically adjusts the number of retained candidate nodes based on channel conditions. With the ability to evaluate channel status dynamically, the proposed method achieves low-complexity decoding and enhances transmission performance. In simulations under channel conditions with SNR ranging from -14 to 5 dB, the average decoding complexity was reduced by 58% compared to conventional FSD. In communication experiments using a dynamically variable optical attenuator to generate random attenuation between 0 and -30 dB, the average decoding complexity was reduced by 23.6% compared to FSD. Furthermore, this work integrates the non-quantized forward stack decoding with the superposition UEP-Spinal code to construct a novel MWD communication system. In simulated MWD environment, the proposed system achieved an average bit error rate (BER) of $5.58\times 10^{-5}$ , with BER consistently below 0.1% across all intervals. Compared to the traditional Spinal code system, the decoding complexity was reduced by 31.5%. The results demonstrate that the newly developed communication system exhibits high reliability, low decoding latency, and high effective transmission rates in simulated drilling platform scenarios, making it well-suited for MWD communication applications.